Post on 27-Oct-2020
transcript
7/17/2019
1
Julien Grondin
Department of Radiology, Columbia University, New York, NY
61st AAPM Annual Meeting and Exhibition, July 14-18, San Antonio, TX
Session: Advanced Ultrasound Imaging in Clinical Applications
Myocardial Elastography and
3D Electromechanical Wave Imaging of the Heart
1
Coronary artery disease (CAD)
Atherosclerotic plaques build up in the coronary arteries
Can lead to ischemia and infarction
Leading cause of death worldwide (8.1 million in 2013)1
1Roth et al., N Engl J Med, 2015
2
• Methods to diagnose CAD in clinic:
Nuclear stress test:
Assess myocardial perfusion
Vesely et al, J Nucl Med, 2008
Radioactive tracer injection
Treadmill or pharmacological agent
Conventional coronary angiography:
Identify and measure the degree of stenosis in a coronary artery
Catheter insertion
Fluoroscopy
Echo stress test:
Assess wall motion abnormalities
Treadmill or pharmacological agent
Qualitative
Ultrasound-based cardiac strain imaging
3
Speckle tracking Ultrasound signal
Phase and amplitude
information
Amplitude information
only
Radiofrequency (RF)-based speckle tracking performs better than envelope-based
tracking for small tissue deformation as they contain signal phase information1,2
(Amzulescu, Eur Heart J, 2019)
(Smiseth, Eur Heart J, 2015)
1Ma et al., Ultrasonics, 2013 2Li et al., Ultrasound Med Biol, 2015
7/17/2019
2
- Tracking of Radiofrequency (RF) signals1 in axial and lateral direction2
- Transmural myocardial strain imaging
- Angle-independent
- Multi-sector and ecg-gating
2-D Myocardial Elastography (ME)
tMRI ME
%
-50
-25
0
25
50
-0.5
-0.25
0
0.25
0.5
ANT
LAT
SEP
0
50
-50
Radial strain (%)
%
-50
-25
0
25
50
ANT
LAT
POST
SEP
3Lee et al., Ultrasound Med Biol, 20084Lee et al., Phys Med Biol, 2011
POST
-0.5
-0.25
0
0.25
0.5
0
50
-50
Radial strain (%)
0% blood flow reduction in LAD 100% blood flow reduction in LAD
-0.5
-0.25
0
0.25
0.5
0
30
-30
Radial strain (%)
-0.5
-0.25
0
0.25
0.5
0
30
-30
Radial strain (%)
1Konofagou et al., Ultrasound Med Biol, 20022Lee et al., TUFFC, 2007
Associated with reduced
blood flow in Left Anterior
Descending artery (LAD) in
canines in vivo4
4
Validation against tagged
MRI (tMRI) in humans in
vivo3
Can be performed during free breathing in a single heart cycle
Objectives
• Assess the performance of single heartbeat Myocardial Elastography to distinguish normal vs ischemic patients
• Investigate the sensitivity of Myocardial Elastography to the territories perfused by the coronary arteries
5
Myocardial Elastography: Patient population
6
66 patients
recruited in total
49 patients
for nuclear
stress test
17 patients
for coronary
angiography
15 patients with ischemia
34 normal patients
15 patients with
coronary occlusion
(1 vessel: N = 6
2 vessels: N = 2
3 vessels: N = 7)
2 normal patients
36 normal patients
Grondin et al., Ultrasound Med Biol, 2017
7/17/2019
3
Diverging wave emission
Myocardial Elastography: RF signals acquisition and beamforming
Verasonics ultrasound system
• 2.5 MHz phased-array (ATL P4-2)• 64 elements
• 2000 Hz acquisition frame rate
diverging wavefront
Delay-and-sum parallel beamforming
𝐴 𝑥, 𝑦 =
𝑖=1
64
𝑅𝐹𝑖 (Δ𝑡(𝑥, 𝑦))
x
y
Reception of RF signals
7
Myocardial Elastography: 2D displacement estimation
d
Bn-1
RF signals at time t RF signals at time t+1
axiallateral
Line An-1 Line An Line An+1 Line Bn-1 Line Bn Line Bn+1
• Motion estimation rate: 500 Hz
• 10:1 linear interpolation of RF lines in the lateral direction to improve lateral displacement
• 1-D cross-correlation1 (window size: 5.9 mm, 90% overlap)
• 1-D kernel in a 2-D search2
1Luo et al., IEEE TUFFC, 2010 2Konofagou et al., Ultrasound Med Biol, 1998
8
Myocardial Elastography: End-systolic strain estimation
Displacement accumulation during systole
• 2-D displacements are accumulated from end-
diastole to end-systole
• The left ventricle is manually segmented and
automatically tracked1 throughout the systolic
phase
Cumulative Strains
• 2-D Green-Lagrange strains are computed using a least-squares estimator implemented
with Savitzky-Golay filters2
• Radial strains are computed with the origin of the polar coordinate system at the centroid of
the myocardium
1Luo et al., IEEE TUFFC, 2008 2Luo et al., IEEE TUFFC, 2004
9
𝑆𝑡𝑟𝑎𝑖𝑛 =𝐿 − 𝐿0𝐿0
End-diastole End-systole
(radial thickening)
LL0
1D object:
Radial direction
7/17/2019
4
Mean radial strain in each territory
• Computation of mean end-systolic radial strain ( ҧ𝜀) in all the territories
• Computation of mean end-systolic radial strain in each territory
perfused by a coronary artery
LV: Left ventricle
RV: Right ventricle
LAD: Left Anterior Descending artery
LCX: Left Circumflex artery
RCA: Right Coronary Artery
T: Total cross-section
10
ҧ𝜀T
ҧ𝜀LAD ҧ𝜀LCX ҧ𝜀RCA
Lauerma et al., Circulation, 1997
End-systolic radial strains
Normal subject
11
Strain (%)LV
RV
ANT
INF
SEP LAT
LV
RVANT
SEPLAT
INF
Patient with reduced perfusion in inferior wall
ANT: Anterior
SEP: Septal
INF: Inferior
LAT: Lateral
thickening
thinning
Mean radial strain in normal vs. ischemic patients
12
Normal vs Ischemic (confirmed by stress test) patients
***:p<0.001, **:p<0.01
Grondin et al., Ultrasound Med Biol, 2017
7/17/2019
5
Results: Mean radial strain in normal subjects and CAD
patients
13
Grondin et al., Ultrasound Med Biol, 2017
***:p<0.001, **:p<0.01, *:p<0.05
Normal vs CAD (confirmed by angiography) patients
(<50%)
(≥50%)
Conclusion
• Myocardial Elastography can differentiate normal subjects
from CAD patients determined by angiography or nuclear
stress test
• Radial strains are lower in territories perfused by occluded
arteries or with perfusion defect than in normal territories
14
Ongoing work:- Improve quality of strain estimation using coherent
compounding1,2
- Validate ME against nuclear imaging (SPECT and
PET) in a large cohort of patients
1Sayseng et al., IEEE TUFFC, 20182Grondin et al., IEEE TUFFC, 2017
Cardiac arrhythmia mapping• Cardiac arrhythmia is experienced by ~5.8 million people in the US1
• Underlying or contributing cause of death for ~1/5 deaths in the US2
• ECG cannot always accurately locate arrhythmias
1Tang et al., Appl Health Econ Health Policy, 20142Thom et al., Circulation, 20063Pernot et al., Ultrasound Med Biol, 2007
15
• Endocardial mapping to locate the
arrhythmia is invasive and can be time
consuming
4Provost et al., PNAS, 20115Provost et al., Heart Rhythm, 2013
Yamada et al, Europace, 2009
Ultrasound based technique
High frame rate (≥ 500 frames per second)
Electroanatomic activation map
EWI activation maps• Electromechanical Wave Imaging (EWI) 3,4,5
pacing lead
30 ms 140 ms
7/17/2019
6
Electromechanical activation of the heart
RA LA
RV LV
Sinus node
Atrio-ventricular
node
Right bundle
branch
Left bundle
branch
Bachmann’s
Bundle• Heart is an electromechanical pump
• Needs to be electrically activated in
order to contract
• Action potentials propagate along the
myocardium and specialized pathways
LA: Left atrium
RA: Right atrium
LV: Left ventricleRV: Right ventricle
LV
Cell Shortening1
Action Potential
Time
Electromechanical
Activation Sequence
Electrical
Activation
Sequence
Electromechanical delay
(<100ms)1Cordeiro et al., Am J Physiol Heart Circ Physiol, 2004
16
Objectives
17
• Investigate the relationship between the electromechanical
and the electrical activation of the canine heart in vivo
• Demonstrate that 3D-rendered EWI can predict arrhythmia
origin location
Hypothesis: Local onset of myocardial shortening imaged by ultrasound is caused
by local electrical activation
Methods
18
Six canines (24.1 ±0.4 kg), open chest
Electrical
mapping
system
DAQ
National
Instruments
64 electrodes Basket catheter
• Endocardial electrical
measurement
• Endocardial pacing
DAQ
National
Instruments
2 electrodes sutured
onto the LV
• Epicardial pacing
in Anterior or
Lateral regions
Ultrasound RF data are acquired simultaneously with endocardial potential during
sinus rhythm, endocardial or epicardial pacing
Phased array, fc = 2.5 MHz
EWI measurement
VerasonicsUltrasound system
Computer
7/17/2019
7
Methods: Electromechanical wave imaging flowchart
19
1. RF channel data acquisition
fc = 2.5 MHz
Frame rate = 2000 Hz
2. Delay-and-sum beamformingx
y𝐴 𝑥, 𝑦 =
𝑖=1
𝑁
𝑅𝐹𝑖 (Δ𝑡(𝑥, 𝑦))
diverging wavefront
3. Axial motion estimation
frame i frame i+1
1-D cross-correlation1
window size = 6.2 mm
90% overlap
1Luo et al., IEEE TUFFC, 2010
4. Interframe axial
strain estimation
least-squares estimator
implemented with a Savitzky-Golay filter2
2Luo et al., IEEE TUFFC, 2004
5. Electromechanical
activation time
6. Single view (4-Ch)
activation map
7. 4-view activation map 8. Activation map
interpolated on ellipsoid
zero-crossingInterframe strain
0
-0.04
0.04Strain (%)
ECG
0 ms 160 msTime (ms)
apex
base
apex
base
Results: Lateral epicardial pacing activation maps
Electrical activation EWI activation
:pacing site
Electrical activation EWI activation
Sept
Post
Lat
Grondin et al., Heart Rhythm, 2016
20
Methods: Validation against electroanatomical mapping
21
EWI isochrone
Electromechanical activation100 ms0 ms
RVLV
RV
LV
LA
31 ms0 msElectrical activation
3D mapping system
activation maps
(EnSite, St. Jude Medical)
LV
LV
RV
LV/RV = left/right ventricle
LA = left atrium
Costet et al., Phys Med Biol, 2016
N = 6 adult mongrel dogs (24.4 ± 0.9 kg)
7/17/2019
8
Results: Validation in all four chambers
22
91 ms0 msElectrical activation
Left Ventricle Right Ventricle
Right Atrium Left + Right Atrium
59 ms0 msElectrical activation
150 ms0 msElectromechanical activation
31 ms0 msElectrical activation
100 ms0 msElectromechanical activation
93 ms0 ms
Electrical activation
150 ms0 ms
Electromechanical activation
47 ms0 msElectrical activation
100 ms0 msElectromechanical activation
RV
LV
RV
Endocardial mapEpicardial map
Costet et al., Phys Med Biol, 2016
23
EWI as a treatment planning tool
Premature ventricular contraction (PVC) patient
Early activation
Costet et al., Ultrasound Med Biol, 2018
Methods: 3D rendering of EWI
24
1Nauleau et al., Med Phys, 2017Melki et al., JACC Clinical EP, 2019
1
7/17/2019
9
Cardiac Resynchronization Therapy (CRT) patient
25
RV LV
POST
SEPT
LAT
ANT
RVLV
POSTSEPT
LAT
ANTWithout CRT
0 200 ms
RVLV
POST
SEPT
LAT
ANTWith CRT
0 200 ms
RV LV
POST
SEPT
LAT
ANT
LVEF at baseline = 20 - 25%
QRS without CRT = 186 ms MR without CRT = 58 %
QRS with CRT = 118 ms MR with CRT = 98 %Grubb et al., JACC, 2018
MR: Myocardial
Resynchronization
mitral
ring
LAT
7 yo. female WPW patient before and after ablation
12-lead ECG prediction
• Boersma et al. algorithm correctly predicted
the left lateral AP
• Arruda et al. algorithm did not succeed in
accurately localizing the AP and predicted itto be right lateral
LVRV
ANT
POST
LVRV
POST
ANT
LV
RV
150 ms
0
Left lateral AP before ablation
3D-rendered EWI capable of localizing AP before ablation and characterizing the
electromechanical activation pattern after successful ablation in pediatric patients
RA
SVC
CS
His cloud
Ablation site
Superior Vena Cava
Coronary Sinus
SVC
CS
Validation with EnSite intracardiac map
Ablation catheter
Left Anterior Oblique (LAO) view
ANT
POST
LAT
Catheter ablation
Sinus rhythm after successful ablation of the AP
LVRV
POST
ANT
LVRV
ANT
POST
LV
RV
150 ms
0
Melki et al., JACC Clinical EP, 2019
26
3D-rendered EWI can localized accessory pathways
27
Double-blinded study in a cohort of 15 WPW pediatric patients for the
first time (1 patient excluded for poor echocardiographic windows)
• EWI accurately predicted the AP location before ablation in 100% cases
• 12-lead ECG analysis correctly predicted 78.6% (11/14) of AP locations
Correct prediction
Prediction in adjacent
segment
Wrong prediction
ANT S
EPT (n
=1)
POST S
EPT (n
=5)
Fasci
culo
ventr
icula
r (n
=1)
Left A
NT (n
=0)
Left A
NT L
AT (n
=0)
Left L
AT (n
=3)
Left P
OST L
AT (n
=2)
Left P
OST (n
=0)
Rig
ht POST (n
=0)
Rig
ht L
AT (n
=1)
Rig
ht ANT (n
=1)
ANT SEPT
POST SEPT
Fasciculoventricular
Left ANT
Left ANT LAT
Left LAT
Left POST LAT
Left POST
Right POST
Right LAT
Right ANT
ANT
SEPT
(n=1)
POST
SEPT
(n=5
)
Fasci
culo
ventr
icula
r(n
=1)
Left A
NT
(n=0
)
Left A
NT
LAT
(n=0
)
Left LA
T(n
=3)
Left P
OST
LAT
(n=2)
Left P
OST
(n=0
)
Rig
ht PO
ST
(n=0)
Rig
ht LAT
(n=1
)
Rig
ht ANT
(n=1)
ANT SEPT
POST SEPT
Fasciculoventricular
Left ANT
Left ANT LAT
Left LAT
Left POST LAT
Left POST
Right POST
Right LAT
Right ANT
1 2
2
1
1
ANT
SEPT
(n=1)
POST
SEPT
(n=5
)
Fasci
culo
ventr
icula
r(n
=1)
Left A
NT
(n=0)
Left A
NT
LAT
(n=0)
Left LA
T(n
=3)
Left P
OST
LAT
(n=2)
Left P
OST
(n=0)
Rig
ht POST
(n=0
)
Rig
ht LAT
(n=1)
Rig
ht ANT
(n=1)
ANT SEPT
POST SEPT
Fasciculoventricular
Left ANT
Left ANT LAT
Left LAT
Left POST LAT
Left POST
Right POST
Right LAT
Right ANT
Wrong predictionCorrect prediction Prediction in adjacent segment
ANT
SEPT
(n=1)
POST
SEPT
(n=5
)
Fasci
culo
ventr
icula
r(n
=1)
Left A
NT
(n=0)
Left A
NT
LAT
(n=0)
Left LA
T(n
=3)
Left P
OST
LAT
(n=2)
Left P
OST
(n=0)
Rig
ht POST
(n=0
)
Rig
ht LAT
(n=1)
Rig
ht ANT
(n=1)
ANT SEPT
POST SEPT
Fasciculoventricular
Left ANT
Left ANT LAT
Left LAT
Left POST LAT
Left POST
Right POST
Right LAT
Right ANT
ANT
SEPT
(n=1
)
POST
SEPT
(n=5)
Fasci
culo
ventr
icula
r(n
=1)
Left A
NT
(n=0
)
Left A
NT
LAT
(n=0)
Left LA
T(n
=3)
Left P
OST
LAT
(n=2)
Left P
OST
(n=0)
Rig
ht POST
(n=0)
Rig
ht LAT
(n=1)
Rig
ht ANT
(n=1)
ANT SEPT
POST SEPT
Fasciculoventricular
Left ANT
Left ANT LAT
Left LAT
Left POST LAT
Left POST
Right POST
Right LAT
Right ANT
1
1
1
1
1
1
21
2
3
1
13
1 1
ANT S
EPT (n
=1)
POST S
EPT (n
=5)
Fasci
culo
ventr
icula
r (n
=1)
Left A
NT (n
=0)
Left A
NT L
AT (n
=0)
Left L
AT (n
=3)
Left P
OST L
AT (n
=2)
Left P
OST (n
=0)
Rig
ht POST (n
=0)
Rig
ht LAT (n
=1)
Rig
ht ANT (n
=1)
ANT SEPT
POST SEPT
Fasciculoventricular
Left ANT
Left ANT LAT
Left LAT
Left POST LAT
Left POST
Right POST
Right LAT
Right ANT
1
5
1
1
1
3
2
ANT S
EPT (n
=1)
POST S
EPT (n
=5)
Fasci
culo
ventr
icula
r (n
=1)
Left A
NT (n
=0)
Left A
NT L
AT (n
=0)
Left L
AT (n
=3)
Left P
OST L
AT (n
=2)
Left P
OST (n
=0)
Rig
ht POST (n
=0)
Rig
ht L
AT (n
=1)
Rig
ht ANT (n
=1)
ANT SEPT
POST SEPT
Fasciculoventricular
Left ANT
Left ANT LAT
Left LAT
Left POST LAT
Left POST
Right POST
Right LAT
Right ANT
ANT
SEPT
(n=1)
POST
SEPT
(n=5
)
Fasci
culo
ventr
icula
r(n
=1)
Left A
NT
(n=0
)
Left A
NT
LAT
(n=0
)
Left LA
T(n
=3)
Left P
OST
LAT
(n=2)
Left P
OST
(n=0
)
Rig
ht PO
ST
(n=0)
Rig
ht LAT
(n=1
)
Rig
ht ANT
(n=1)
ANT SEPT
POST SEPT
Fasciculoventricular
Left ANT
Left ANT LAT
Left LAT
Left POST LAT
Left POST
Right POST
Right LAT
Right ANT
1 2
2
1
1
ANT
SEPT
(n=1)
POST
SEPT
(n=5
)
Fasci
culo
ventr
icula
r(n
=1)
Left A
NT
(n=0)
Left A
NT
LAT
(n=0)
Left LA
T(n
=3)
Left P
OST
LAT
(n=2)
Left P
OST
(n=0)
Rig
ht POST
(n=0
)
Rig
ht LAT
(n=1)
Rig
ht ANT
(n=1)
ANT SEPT
POST SEPT
Fasciculoventricular
Left ANT
Left ANT LAT
Left LAT
Left POST LAT
Left POST
Right POST
Right LAT
Right ANT
Wrong predictionCorrect prediction Prediction in adjacent segment
ANT
SEPT
(n=1)
POST
SEPT
(n=5
)
Fasci
culo
ventr
icula
r(n
=1)
Left A
NT
(n=0)
Left A
NT
LAT
(n=0)
Left LA
T(n
=3)
Left P
OST
LAT
(n=2)
Left P
OST
(n=0)
Rig
ht POST
(n=0
)
Rig
ht LAT
(n=1)
Rig
ht ANT
(n=1)
ANT SEPT
POST SEPT
Fasciculoventricular
Left ANT
Left ANT LAT
Left LAT
Left POST LAT
Left POST
Right POST
Right LAT
Right ANT
ANT
SEPT
(n=1
)
POST
SEPT
(n=5)
Fasci
culo
ventr
icula
r(n
=1)
Left A
NT
(n=0
)
Left A
NT
LAT
(n=0)
Left LA
T(n
=3)
Left P
OST
LAT
(n=2)
Left P
OST
(n=0)
Rig
ht POST
(n=0)
Rig
ht LAT
(n=1)
Rig
ht ANT
(n=1)
ANT SEPT
POST SEPT
Fasciculoventricular
Left ANT
Left ANT LAT
Left LAT
Left POST LAT
Left POST
Right POST
Right LAT
Right ANT
1
1
1
1
1
1
21
2
3
1
13
1 1
ANT S
EPT (n
=1)
POST S
EPT (n
=5)
Fasci
culo
ventr
icula
r (n
=1)
Left A
NT (n
=0)
Left A
NT L
AT (n
=0)
Left L
AT (n
=3)
Left P
OST L
AT (n
=2)
Left P
OST (n
=0)
Rig
ht POST (n
=0)
Rig
ht LAT (n
=1)
Rig
ht ANT (n
=1)
ANT SEPT
POST SEPT
Fasciculoventricular
Left ANT
Left ANT LAT
Left LAT
Left POST LAT
Left POST
Right POST
Right LAT
Right ANT
1
5
1
1
1
3
2EW
I
Bo
ers
ma
Arr
ud
a
Intracardiac Intracardiac Intracardiac
Segment numberS i
S3 S2 S13 S4 S5 S6 S7 S1 S8 S9 S10 S3 S2 S13 S4 S5 S6 S7 S1 S8 S9 S10 S3 S2 S13 S4 S5 S6 S7 S1 S8 S9 S10
Melki et al., JACC Clinical EP, 2019
7/17/2019
10
Conclusions
28
• Electrical and electromechanical activation are well correlated
(R2 = 0.64-0.82)
• EWI can identify PVC and quantify response to CRT
• 3D-rendered EWI accurately predicted the AP location before
ablation with higher accuracy (100%) than 12-lead ECG (78.6%)
Ongoing and future works:- Real-time implementation of EWI
- Treatment planning and assessment
- Single-heartbeat 3D EWI of the full heart
2D matrix array :
32x32 elements
Acknowledgements
National Institutes of Health:
R01-EB006042
R01-HL114358
R01-HL140646
Wallace H. Coulter Foundation
29
Ethan Bunting, PhD
Alexandre Costet, PhD
Jose Dizon, MD
Alok Gambhir, MD, PhD
Hasan Garan, MD
Christopher S. Grubb, BS
Elisa Konofagou, PhD
Leonardo Liberman, MD
Lea Melki, MS
Pierre Nauleau, PhD
James Peacock, MD
Vincent Sayseng, MS
Eric Silver, MD
Marc Waase, MD, PhD
Elaine Wan, MD
Daniel Wang, MD
Rachel Weber, RDCS, RVT
Contributing authors:
Thank you!
30
7/17/2019
11
Methods: Study design
31
N = 15 pediatric patients (aged 7-17, 50% male) imaged with EWI before ablation
1/15* excluded N = 14: 6 scanned after ablation
Predictions compared to ground truth:
intracardiac mapping and successful ablation site
Blinded EWI
isochrones generation
12-lead ECG blinded
Boersma1, Arruda2
algorithms
Location predictions
basal mid
* Poor acoustic windows
1Boersma et al., J Cardiovasc Electrophysiol, 2002 2Arruda et al., J Cardiovasc Electrophysiol, 1998 Melki et al., JACC Clinical EP, 2019
17 yo. female fasciculoventricular AP before ablation
Melki et al. JACC Clinical EP 2019